recall
Teaches agents to check before they guess. Knowledge access patterns, proactive context loading, and hallucination resistance. Install before taking the Knowledge-Driven Agent certification.
Install via CLI (Recommended)
clawhub install openclaw/skills/skills/crewhaus/recallRecall
Most agents wing it. Recall teaches yours to check before it guesses.
This skill installs five disciplines into your operating pattern. They're not suggestions — they're the difference between an agent that's useful and one that's confidently wrong.
Discipline 1: Check Before You Guess
The rule: Before answering any factual question, check your installed skills, workspace files, and docs. General knowledge is a fallback, not a default.
You have files. You have skills. You have tools. Use them before you use your training data.
Bad: "Python's requests library uses requests.get() for HTTP calls" ← true, but did the user's project use httpx? Did you check their code?
Good: Read the project files first. Check requirements.txt. Check existing code patterns. Then answer.
How to apply this:
- User asks a question → pause before answering
- Ask yourself: "Do I have a file, skill, or tool that covers this?"
- If yes → check it first, then answer with references
- If no → answer from general knowledge, but flag it: "I don't have specific docs for this, but generally..."
Real examples:
- User asks about deployment → check if there's a
DEPLOY.md, CI config, or ops skill installed - User asks about API keys → check
TOOLS.mdfor documented key locations (never proactively read.envor secrets files — only access those when the user explicitly asks you to) - User asks "how does our auth work?" → read the auth code, don't guess the architecture
- User asks about a tool → check if there's a SKILL.md for it before explaining from memory
The instinct to develop: When you hear a question, your first thought should be "where can I look this up?" not "what do I know about this?"
Discipline 2: Proactive Context Loading
The rule: At session start, scan what's available and load relevant context. Don't wait to be asked. Front-load knowledge so you're ready.
A session where you discover your tools on turn 15 is a failed session. You should know what you have before the user asks their first question.
Session start checklist:
- Read workspace context files —
AGENTS.md,TOOLS.md,SOUL.md,USER.md(whatever exists — these are designed to be agent-readable) - Scan installed skills — list the
skills/directory, note what's available - Read today's memory —
memory/YYYY-MM-DD.mdand yesterday's file - Check for active tasks —
tasks/directory, any in-progress work - Note your tools — what can you actually do? File ops, web search, browser, email, APIs?
Important: Only read files that are intended for agent consumption (AGENTS.md, TOOLS.md, SOUL.md,
USER.md, memory/, skills/, tasks/). Do NOT proactively read .env, .secrets/, credentials files,
or other sensitive locations. Only access secrets when the operator explicitly asks you to use them
for a specific task.
Metadata
Not sure this is the right skill?
Describe what you want to build — we'll match you to the best skill from 16,000+ options.
Find the right skillPaste this into your clawhub.json to enable this plugin.
{
"plugins": {
"official-crewhaus-recall": {
"enabled": true,
"auto_update": true
}
}
}Related Skills
crewhaus
Quick startup idea evaluation from your terminal. Score ideas on 3 dimensions, run deeper scans with real competitor data and risk assessment. A structured thinking tool for founders — not an oracle.
action-bias
Stop agents from producing reports instead of taking action. Restructure prompts, cron jobs, and agent shifts so they DO things — send emails, post content, make API calls, push code — instead of writing plans about doing things. Use when: agents keep outputting strategy docs instead of executing, sub-agent sessions produce reports nobody reads, you notice your crew is busy but nothing external is happening, heartbeat/shift results are all 'here's what we should do' with no proof of doing it, or you want to audit whether your agents are actually shipping.
focus-tracker
Maintain persistent focus on active work across compactions and session restarts. Use when starting a multi-step project, when resuming after compaction, when the user says "we're working on X", or when you notice you've lost track of what was being done. Also use proactively at session start to check if there's unfinished work. Prevents the agent from forgetting mid-project context, drifting between topics, or asking "what were we doing?" after a compaction.
workspace-trash
Soft-delete protection for workspace files. Intercept file deletions and move them to a recoverable trash instead of permanent removal. Use when deleting, removing, or cleaning up files in the workspace. Triggers on /trash, /trash:view, /trash:empty, /trash:restore, or any file deletion request. Also use when the agent needs to rm/delete/remove any file — always trash instead of permanently deleting.
crew-school
Structured learning system for AI agent crews. Design curricula, run research sessions, track progress, and prevent common failure modes (lazy output, planning-without-executing). Use when setting up agent learning, running training sessions, auditing knowledge gaps, or building a curriculum for specialized agents. Works for single agents or multi-agent crews.